Pnc Software Engineer Interview Guide

The main round was a Kafka deep-dive, and it went deeper than I expected. It started comfortable — producers, consumers, partitioning strategy, stuff I have lived in for years. Then the interviewer started pulling threads: what actually happens during a consumer group rebalance, what breaks mid-flight, how do you guarantee you do not process the same payment event twice. That is where I started sweating a little — not because I did not know it, but because they wanted the failure-mode version, not the textbook version. I walked through idempotent deduplication the way I had actually built it — dedup keys, exactly-once semantics and where "exactly-once" is honestly marketing — and I could see the interviewer relax once I admitted the messy parts instead of pretending Kafka is clean. The surprise was how much they cared about testing: I ended up describing how I would structure an automated Kafka test framework, which turned into almost a whiteboarding session. Confident on the fundamentals, sweaty on the edge cases, and honestly the edge-case sweat is what got me through.

The specific questions covered: which Spring Kafka annotations apply on the consumer vs producer side; whether any annotation matters on the producer (they wanted @Transactional backed by KafkaTransactionManager); how to maintain message sequence on a hot partition after a rebalance (key-based partitioning, ConsumerRebalanceListener, manual offset commits with MANUAL_IMMEDIATE ack mode); at-least-once vs exactly-once semantics and idempotent consumer design; retry strategy and dead letter topic routing; consumer lag detection and per-partition lag metrics. Behavioral questions included the hardest production bug I had dealt with and ordering guarantees in financial event streams.